Extremity Game
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/0xnurl/signaling-auto-encoder
下载链接
链接失效反馈官方服务：
资源简介：
该数据集旨在激励并测试丰富的组合性，其中对象由可度量的属性定义，信息可能包含组合性要素。该数据集被用于分析神经代理在交流中的成功程度以及离散信息的出现。在严格环境下，数据集包含10个对象，而在非严格环境下，对象数量可以是5、10或15个。该任务通过神经代理的信号游戏来研究语言的涌现。
This dataset is designed to inspire and test rich compositionality, where objects are defined by measurable attributes and the information may contain compositional elements. It is utilized to analyze the success of neural agents in communication and the emergence of discrete information. In the strict setting, the dataset contains 10 objects, while in non-strict settings, the number of objects can be 5, 10, or 15. This task studies the emergence of language through signaling games involving neural agents.
